Flood Optimization

Introduction


Flooding Optimization allows Open/R to greatly reduce control plane traffic. We use DUAL (diffusing update algorithm) to let Open/R to form a SPT (spanning tree, not necessarily MST) used as a flooding topology. Open/R will only flood KvStore Updates on flooding topology which is a subset of physical topology. This reduces flooding-updates complexity from O(E) down to O(V).

DUAL


code path: openr/dual/

This part is the core DUAL algorithm library. It’s a combination of distance-vector and Diffusing algorithm. To address looping problem in distance-vector, there is a few solutions, one of them being specifying the whole path to make it path-vector (like BGP does), DUAL instead leverage diffusing algorithm to make sure Parent Node won’t take any action until all its dependent nodes have successfully calculated their paths. Here’s few important specs DUAL library carries:

  • Make sure DUAL nodes can form SPT correctly, every other nodes will find shortest path towards the root (similarly to finding shortest path towards prefix as routing protocol does). (NOTE: each DUAL run can end up different SPTs if there is multiple shortest path towards the root, but SPT formation is guaranteed)

  • Make sure DUAL nodes can handle any network failures (e.g any single/multiple node/link failures under any topology can be handled by DUAL properly -> reform new SPT)

  • Convergence time is small (benchmark in simulation 1024 grid topology, takes < 1 sec to converge)

  • Can form multiple SPT if multiple roots are specified.

KvStore Integration


code path: openr/kvstore/

This part integrate DUAL library with KvStore, here’s how it works:

  • KvStore is inheriting from DUAL, so each KvStore node can talk to its neighbor via DUAL protocol

  • KvStore overrides I/O part via zmq currently (can be replaced by others like thrift)

  • Whenever a node’s nexthop towards root got changed (due to network event e.g node/link flap etc), it will synchronize with its old and new parent to make sure SPT information is consistent on both ends.

  • Multiple KvStore nodes can be picked as roots, in that case, multiple SPT trees would be formed but only one of them will be USED as flooding topology based on node’s priority (currently smallest node’s name would win). If current root got restarted, SPT formed by 2nd priority root will become current flooding topology.

  • When a node needs to flood its update, it will only flood to its dual-parent and dual-children instead of all neighbors as before.

  • In case an update needs to be flooded while DUAL node is not ready (e.g it’s in the process of forming SPT, we call it ACTIVE mode), in this mode, KvStore will fall back to legacy flooding mode by flooding everywhere. Once DUAL is done, it goes back to PASSIVE mode, flooding optimization will take effect again.

  • Handle parallel link cases as well. From DUAL perspective, parallel link will be treated as a single neighbor.

  • Backward compatible. You can have a single open/r domain, where 30% is flooding disabled, 70% enabled. 70% nodes would leverage DUAL to perform flooding optimization while 30% will perform legacy flooding.